A minimal AI system:
It must be of a certain size to percieve a domain of a certain size:  a 40 bit system can’t understand a billion bits.  There must be some information limit to the size of brain capacity to understand something of a certain complexity.

There must be robust set of inputs that provide lots of data to the system

There must be a robust ability to interact with the environment so the system can cause action and then see results to validate generalizations made

The system requires a powerful pattern matching scheme

The system requires a powerful generalization mechanism and ability to correct bad generalizations, unlearn them

The system requires that generalizations be made close to the source and then processed automatically so that only the generalizations pass up or that the generalizations with exceptions are passed up (this is like x but differs in y and z)

The system requires a lot of memory of sequences of generalizations and specific data that can be recalled

The system requires being able to link different generalizations with other generalizations and other inputs and these can be used to link together and form other generalizations

The system requires a motivation to do anything and to self-correct

The system may require active teaching because it may require lessons planned in advance and ways of testing if it is producing good results to facilitate higher learning

The problems to try to explain:

dreams that seem planned in advance

how ideas are formed in dreams to solve problems in real life

how “aha” moments happen

How can the brain learn to do things like complex physical activities that seem to take small fractions of a second and integration of various inputs be processed in milliseconds in coordination with physical activity that is fine tuned

Does the body itself learn without the brain?

What kinds of metrics can be applied to understand the scope of the calculation, the pattern matching required to do an activity x